KIS Trading System and TinyGPT
An ECO4126 project combining safe trading-system simulation with a small decoder-only Transformer.
Overview
This coursework repository contains two applied software projects: a simulation-first automated trading system designed around the Korea Investment & Securities Open API concept, and TinyGPT, a compact GPT-2-style language model implemented in PyTorch.
KIS trading-system component
- Organizes the workflow as market data, strategy, risk management, order execution, and logging.
- Uses a moving-average crossover strategy to generate buy, sell, or hold signals.
- Applies position-size, daily-trade, stop-loss, and take-profit controls.
- Runs in simulation mode by default to prevent unintended real orders and credential exposure.
TinyGPT component
- Implements token and positional embeddings, masked multi-head self-attention, feed-forward layers, residual connections, and layer normalization.
- Trains with character-level next-token prediction and cross-entropy loss.
- Supports checkpointing and autoregressive text generation.
Technologies: Python, PyTorch, Jupyter Notebook, Transformer architecture, API-oriented system design, and risk controls.